Line data Source code
1 : // SPDX-License-Identifier: GPL-2.0-or-later
2 : /*
3 : * Copyright (C) 2021-2023 Oracle. All Rights Reserved.
4 : * Author: Darrick J. Wong <djwong@kernel.org>
5 : */
6 : #include "xfs.h"
7 : #include "xfs_fs.h"
8 : #include "xfs_shared.h"
9 : #include "xfs_format.h"
10 : #include "xfs_log_format.h"
11 : #include "xfs_trans_resv.h"
12 : #include "xfs_bit.h"
13 : #include "xfs_sb.h"
14 : #include "xfs_mount.h"
15 : #include "xfs_defer.h"
16 : #include "xfs_inode.h"
17 : #include "xfs_trans.h"
18 : #include "xfs_alloc.h"
19 : #include "xfs_btree.h"
20 : #include "xfs_btree_staging.h"
21 : #include "xfs_rtrefcount_btree.h"
22 : #include "xfs_refcount.h"
23 : #include "xfs_trace.h"
24 : #include "xfs_cksum.h"
25 : #include "xfs_error.h"
26 : #include "xfs_extent_busy.h"
27 : #include "xfs_rtgroup.h"
28 : #include "xfs_rtbitmap.h"
29 : #include "xfs_imeta.h"
30 : #include "xfs_health.h"
31 :
32 : static struct kmem_cache *xfs_rtrefcountbt_cur_cache;
33 :
34 : /*
35 : * Realtime Reference Count btree.
36 : *
37 : * This is a btree used to track the owner(s) of a given extent in the realtime
38 : * device. See the comments in xfs_refcount_btree.c for more information.
39 : *
40 : * This tree is basically the same as the regular refcount btree except that
41 : * it's rooted in an inode.
42 : */
43 :
44 : static struct xfs_btree_cur *
45 2789444 : xfs_rtrefcountbt_dup_cursor(
46 : struct xfs_btree_cur *cur)
47 : {
48 2789444 : struct xfs_btree_cur *new;
49 :
50 2789444 : new = xfs_rtrefcountbt_init_cursor(cur->bc_mp, cur->bc_tp,
51 : cur->bc_ino.rtg, cur->bc_ino.ip);
52 :
53 : /* Copy the flags values since init cursor doesn't get them. */
54 2789460 : new->bc_ino.flags = cur->bc_ino.flags;
55 :
56 2789460 : return new;
57 : }
58 :
59 : STATIC int
60 21593988 : xfs_rtrefcountbt_get_minrecs(
61 : struct xfs_btree_cur *cur,
62 : int level)
63 : {
64 21593988 : if (level == cur->bc_nlevels - 1) {
65 168037 : struct xfs_ifork *ifp = xfs_btree_ifork_ptr(cur);
66 :
67 168040 : return xfs_rtrefcountbt_maxrecs(cur->bc_mp, ifp->if_broot_bytes,
68 168040 : level == 0) / 2;
69 : }
70 :
71 21425951 : return cur->bc_mp->m_rtrefc_mnr[level != 0];
72 : }
73 :
74 : STATIC int
75 4874221069 : xfs_rtrefcountbt_get_maxrecs(
76 : struct xfs_btree_cur *cur,
77 : int level)
78 : {
79 4874221069 : if (level == cur->bc_nlevels - 1) {
80 1413932833 : struct xfs_ifork *ifp = xfs_btree_ifork_ptr(cur);
81 :
82 1413556515 : return xfs_rtrefcountbt_maxrecs(cur->bc_mp, ifp->if_broot_bytes,
83 : level == 0);
84 : }
85 :
86 3460288236 : return cur->bc_mp->m_rtrefc_mxr[level != 0];
87 : }
88 :
89 : /*
90 : * Calculate number of records in a realtime refcount btree inode root.
91 : */
92 : unsigned int
93 0 : xfs_rtrefcountbt_droot_maxrecs(
94 : unsigned int blocklen,
95 : bool leaf)
96 : {
97 4012368 : blocklen -= sizeof(struct xfs_rtrefcount_root);
98 :
99 0 : if (leaf)
100 3308993 : return blocklen / sizeof(struct xfs_refcount_rec);
101 678482 : return blocklen / (2 * sizeof(struct xfs_refcount_key) +
102 : sizeof(xfs_rtrefcount_ptr_t));
103 : }
104 :
105 : /*
106 : * Get the maximum records we could store in the on-disk format.
107 : *
108 : * For non-root nodes this is equivalent to xfs_rtrefcountbt_get_maxrecs, but
109 : * for the root node this checks the available space in the dinode fork so that
110 : * we can resize the in-memory buffer to match it. After a resize to the
111 : * maximum size this function returns the same value as
112 : * xfs_rtrefcountbt_get_maxrecs for the root node, too.
113 : */
114 : STATIC int
115 4010553 : xfs_rtrefcountbt_get_dmaxrecs(
116 : struct xfs_btree_cur *cur,
117 : int level)
118 : {
119 4010553 : if (level != cur->bc_nlevels - 1)
120 23078 : return cur->bc_mp->m_rtrefc_mxr[level != 0];
121 7974950 : return xfs_rtrefcountbt_droot_maxrecs(cur->bc_ino.forksize, level == 0);
122 : }
123 :
124 : STATIC void
125 12382583719 : xfs_rtrefcountbt_init_key_from_rec(
126 : union xfs_btree_key *key,
127 : const union xfs_btree_rec *rec)
128 : {
129 12382583719 : key->refc.rc_startblock = rec->refc.rc_startblock;
130 12382583719 : }
131 :
132 : STATIC void
133 205769858 : xfs_rtrefcountbt_init_high_key_from_rec(
134 : union xfs_btree_key *key,
135 : const union xfs_btree_rec *rec)
136 : {
137 205769858 : __u32 x;
138 :
139 205769858 : x = be32_to_cpu(rec->refc.rc_startblock);
140 205769858 : x += be32_to_cpu(rec->refc.rc_blockcount) - 1;
141 205769858 : key->refc.rc_startblock = cpu_to_be32(x);
142 205769858 : }
143 :
144 : STATIC void
145 1107243141 : xfs_rtrefcountbt_init_rec_from_cur(
146 : struct xfs_btree_cur *cur,
147 : union xfs_btree_rec *rec)
148 : {
149 1107243141 : const struct xfs_refcount_irec *irec = &cur->bc_rec.rc;
150 1107243141 : uint32_t start;
151 :
152 1107243141 : start = xfs_refcount_encode_startblock(irec->rc_startblock,
153 1107243141 : irec->rc_domain);
154 1107243141 : rec->refc.rc_startblock = cpu_to_be32(start);
155 1107243141 : rec->refc.rc_blockcount = cpu_to_be32(cur->bc_rec.rc.rc_blockcount);
156 1107243141 : rec->refc.rc_refcount = cpu_to_be32(cur->bc_rec.rc.rc_refcount);
157 1107243141 : }
158 :
159 : STATIC void
160 1780113068 : xfs_rtrefcountbt_init_ptr_from_cur(
161 : struct xfs_btree_cur *cur,
162 : union xfs_btree_ptr *ptr)
163 : {
164 1780113068 : ptr->l = 0;
165 1780113068 : }
166 :
167 : STATIC int64_t
168 15823309800 : xfs_rtrefcountbt_key_diff(
169 : struct xfs_btree_cur *cur,
170 : const union xfs_btree_key *key)
171 : {
172 15823309800 : const struct xfs_refcount_key *kp = &key->refc;
173 15823309800 : const struct xfs_refcount_irec *irec = &cur->bc_rec.rc;
174 15823309800 : uint32_t start;
175 :
176 15823309800 : start = xfs_refcount_encode_startblock(irec->rc_startblock,
177 15823309800 : irec->rc_domain);
178 15823309800 : return (int64_t)be32_to_cpu(kp->rc_startblock) - start;
179 : }
180 :
181 : STATIC int64_t
182 592641212 : xfs_rtrefcountbt_diff_two_keys(
183 : struct xfs_btree_cur *cur,
184 : const union xfs_btree_key *k1,
185 : const union xfs_btree_key *k2,
186 : const union xfs_btree_key *mask)
187 : {
188 592641212 : ASSERT(!mask || mask->refc.rc_startblock);
189 :
190 592641212 : return (int64_t)be32_to_cpu(k1->refc.rc_startblock) -
191 592641212 : be32_to_cpu(k2->refc.rc_startblock);
192 : }
193 :
194 : static xfs_failaddr_t
195 633098 : xfs_rtrefcountbt_verify(
196 : struct xfs_buf *bp)
197 : {
198 633098 : struct xfs_mount *mp = bp->b_target->bt_mount;
199 633098 : struct xfs_btree_block *block = XFS_BUF_TO_BLOCK(bp);
200 633098 : xfs_failaddr_t fa;
201 633098 : int level;
202 :
203 633098 : if (!xfs_verify_magic(bp, block->bb_magic))
204 0 : return __this_address;
205 :
206 633098 : if (!xfs_has_reflink(mp))
207 0 : return __this_address;
208 633098 : fa = xfs_btree_lblock_v5hdr_verify(bp, XFS_RMAP_OWN_UNKNOWN);
209 633098 : if (fa)
210 : return fa;
211 633098 : level = be16_to_cpu(block->bb_level);
212 633098 : if (level > mp->m_rtrefc_maxlevels)
213 0 : return __this_address;
214 :
215 633098 : return xfs_btree_lblock_verify(bp, mp->m_rtrefc_mxr[level != 0]);
216 : }
217 :
218 : static void
219 12041 : xfs_rtrefcountbt_read_verify(
220 : struct xfs_buf *bp)
221 : {
222 12041 : xfs_failaddr_t fa;
223 :
224 12041 : if (!xfs_btree_lblock_verify_crc(bp))
225 0 : xfs_verifier_error(bp, -EFSBADCRC, __this_address);
226 : else {
227 12041 : fa = xfs_rtrefcountbt_verify(bp);
228 12041 : if (fa)
229 0 : xfs_verifier_error(bp, -EFSCORRUPTED, fa);
230 : }
231 :
232 12041 : if (bp->b_error)
233 0 : trace_xfs_btree_corrupt(bp, _RET_IP_);
234 12041 : }
235 :
236 : static void
237 261969 : xfs_rtrefcountbt_write_verify(
238 : struct xfs_buf *bp)
239 : {
240 261969 : xfs_failaddr_t fa;
241 :
242 261969 : fa = xfs_rtrefcountbt_verify(bp);
243 261969 : if (fa) {
244 0 : trace_xfs_btree_corrupt(bp, _RET_IP_);
245 0 : xfs_verifier_error(bp, -EFSCORRUPTED, fa);
246 0 : return;
247 : }
248 261969 : xfs_btree_lblock_calc_crc(bp);
249 :
250 : }
251 :
252 : const struct xfs_buf_ops xfs_rtrefcountbt_buf_ops = {
253 : .name = "xfs_rtrefcountbt",
254 : .magic = { 0, cpu_to_be32(XFS_RTREFC_CRC_MAGIC) },
255 : .verify_read = xfs_rtrefcountbt_read_verify,
256 : .verify_write = xfs_rtrefcountbt_write_verify,
257 : .verify_struct = xfs_rtrefcountbt_verify,
258 : };
259 :
260 : STATIC int
261 198647 : xfs_rtrefcountbt_keys_inorder(
262 : struct xfs_btree_cur *cur,
263 : const union xfs_btree_key *k1,
264 : const union xfs_btree_key *k2)
265 : {
266 198647 : return be32_to_cpu(k1->refc.rc_startblock) <
267 198647 : be32_to_cpu(k2->refc.rc_startblock);
268 : }
269 :
270 : STATIC int
271 108957450 : xfs_rtrefcountbt_recs_inorder(
272 : struct xfs_btree_cur *cur,
273 : const union xfs_btree_rec *r1,
274 : const union xfs_btree_rec *r2)
275 : {
276 108957450 : return be32_to_cpu(r1->refc.rc_startblock) +
277 108957450 : be32_to_cpu(r1->refc.rc_blockcount) <=
278 108957450 : be32_to_cpu(r2->refc.rc_startblock);
279 : }
280 :
281 : STATIC enum xbtree_key_contig
282 0 : xfs_rtrefcountbt_keys_contiguous(
283 : struct xfs_btree_cur *cur,
284 : const union xfs_btree_key *key1,
285 : const union xfs_btree_key *key2,
286 : const union xfs_btree_key *mask)
287 : {
288 0 : ASSERT(!mask || mask->refc.rc_startblock);
289 :
290 0 : return xbtree_key_contig(be32_to_cpu(key1->refc.rc_startblock),
291 0 : be32_to_cpu(key2->refc.rc_startblock));
292 : }
293 :
294 : /* Move the rt refcount btree root from one incore buffer to another. */
295 : static void
296 6370422 : xfs_rtrefcountbt_broot_move(
297 : struct xfs_inode *ip,
298 : int whichfork,
299 : struct xfs_btree_block *dst_broot,
300 : size_t dst_bytes,
301 : struct xfs_btree_block *src_broot,
302 : size_t src_bytes,
303 : unsigned int level,
304 : unsigned int numrecs)
305 : {
306 6370422 : struct xfs_mount *mp = ip->i_mount;
307 6370422 : void *dptr;
308 6370422 : void *sptr;
309 :
310 12740844 : ASSERT(xfs_rtrefcount_droot_space(src_broot) <=
311 : xfs_inode_fork_size(ip, whichfork));
312 :
313 : /*
314 : * We always have to move the pointers because they are not butted
315 : * against the btree block header.
316 : */
317 6370422 : if (numrecs && level > 0) {
318 4648 : sptr = xfs_rtrefcount_broot_ptr_addr(mp, src_broot, 1,
319 : src_bytes);
320 4648 : dptr = xfs_rtrefcount_broot_ptr_addr(mp, dst_broot, 1,
321 : dst_bytes);
322 9296 : memmove(dptr, sptr, numrecs * sizeof(xfs_fsblock_t));
323 : }
324 :
325 6370422 : if (src_broot == dst_broot)
326 : return;
327 :
328 : /*
329 : * If the root is being totally relocated, we have to migrate the block
330 : * header and the keys/records that come after it.
331 : */
332 6316252 : memcpy(dst_broot, src_broot, XFS_RTREFCOUNT_BLOCK_LEN);
333 :
334 3158126 : if (!numrecs)
335 : return;
336 :
337 3005729 : if (level == 0) {
338 3004148 : sptr = xfs_rtrefcount_rec_addr(src_broot, 1);
339 3004148 : dptr = xfs_rtrefcount_rec_addr(dst_broot, 1);
340 6008296 : memcpy(dptr, sptr,
341 : numrecs * sizeof(struct xfs_refcount_rec));
342 : } else {
343 1581 : sptr = xfs_rtrefcount_key_addr(src_broot, 1);
344 1581 : dptr = xfs_rtrefcount_key_addr(dst_broot, 1);
345 3162 : memcpy(dptr, sptr,
346 : numrecs * sizeof(struct xfs_refcount_key));
347 : }
348 : }
349 :
350 : static const struct xfs_ifork_broot_ops xfs_rtrefcountbt_iroot_ops = {
351 : .maxrecs = xfs_rtrefcountbt_maxrecs,
352 : .size = xfs_rtrefcount_broot_space_calc,
353 : .move = xfs_rtrefcountbt_broot_move,
354 : };
355 :
356 : const struct xfs_btree_ops xfs_rtrefcountbt_ops = {
357 : .rec_len = sizeof(struct xfs_refcount_rec),
358 : .key_len = sizeof(struct xfs_refcount_key),
359 : .lru_refs = XFS_REFC_BTREE_REF,
360 : .geom_flags = XFS_BTREE_LONG_PTRS | XFS_BTREE_ROOT_IN_INODE |
361 : XFS_BTREE_CRC_BLOCKS | XFS_BTREE_IROOT_RECORDS,
362 :
363 : .dup_cursor = xfs_rtrefcountbt_dup_cursor,
364 : .alloc_block = xfs_btree_alloc_imeta_block,
365 : .free_block = xfs_btree_free_imeta_block,
366 : .get_minrecs = xfs_rtrefcountbt_get_minrecs,
367 : .get_maxrecs = xfs_rtrefcountbt_get_maxrecs,
368 : .get_dmaxrecs = xfs_rtrefcountbt_get_dmaxrecs,
369 : .init_key_from_rec = xfs_rtrefcountbt_init_key_from_rec,
370 : .init_high_key_from_rec = xfs_rtrefcountbt_init_high_key_from_rec,
371 : .init_rec_from_cur = xfs_rtrefcountbt_init_rec_from_cur,
372 : .init_ptr_from_cur = xfs_rtrefcountbt_init_ptr_from_cur,
373 : .key_diff = xfs_rtrefcountbt_key_diff,
374 : .buf_ops = &xfs_rtrefcountbt_buf_ops,
375 : .diff_two_keys = xfs_rtrefcountbt_diff_two_keys,
376 : .keys_inorder = xfs_rtrefcountbt_keys_inorder,
377 : .recs_inorder = xfs_rtrefcountbt_recs_inorder,
378 : .keys_contiguous = xfs_rtrefcountbt_keys_contiguous,
379 : .iroot_ops = &xfs_rtrefcountbt_iroot_ops,
380 : };
381 :
382 : /* Initialize a new rt refcount btree cursor. */
383 : static struct xfs_btree_cur *
384 491855395 : xfs_rtrefcountbt_init_common(
385 : struct xfs_mount *mp,
386 : struct xfs_trans *tp,
387 : struct xfs_rtgroup *rtg,
388 : struct xfs_inode *ip)
389 : {
390 491855395 : struct xfs_btree_cur *cur;
391 :
392 491855395 : ASSERT(xfs_isilocked(ip, XFS_ILOCK_SHARED | XFS_ILOCK_EXCL));
393 :
394 487273109 : cur = xfs_btree_alloc_cursor(mp, tp, XFS_BTNUM_RTREFC,
395 487273109 : &xfs_rtrefcountbt_ops, mp->m_rtrefc_maxlevels,
396 : xfs_rtrefcountbt_cur_cache);
397 492458642 : cur->bc_statoff = XFS_STATS_CALC_INDEX(xs_refcbt_2);
398 :
399 492458642 : cur->bc_ino.ip = ip;
400 492458642 : cur->bc_ino.allocated = 0;
401 492458642 : cur->bc_ino.flags = 0;
402 492458642 : cur->bc_ino.refc.nr_ops = 0;
403 492458642 : cur->bc_ino.refc.shape_changes = 0;
404 :
405 492458642 : cur->bc_ino.rtg = xfs_rtgroup_hold(rtg);
406 493800313 : return cur;
407 : }
408 :
409 : /* Allocate a new rt refcount btree cursor. */
410 : struct xfs_btree_cur *
411 492425599 : xfs_rtrefcountbt_init_cursor(
412 : struct xfs_mount *mp,
413 : struct xfs_trans *tp,
414 : struct xfs_rtgroup *rtg,
415 : struct xfs_inode *ip)
416 : {
417 492425599 : struct xfs_btree_cur *cur;
418 492425599 : struct xfs_ifork *ifp = xfs_ifork_ptr(ip, XFS_DATA_FORK);
419 :
420 492425599 : cur = xfs_rtrefcountbt_init_common(mp, tp, rtg, ip);
421 494639166 : cur->bc_nlevels = be16_to_cpu(ifp->if_broot->bb_level) + 1;
422 494639166 : cur->bc_ino.forksize = xfs_inode_fork_size(ip, XFS_DATA_FORK);
423 494639166 : cur->bc_ino.whichfork = XFS_DATA_FORK;
424 494639166 : return cur;
425 : }
426 :
427 : /* Create a new rt reverse mapping btree cursor with a fake root for staging. */
428 : struct xfs_btree_cur *
429 28374 : xfs_rtrefcountbt_stage_cursor(
430 : struct xfs_mount *mp,
431 : struct xfs_rtgroup *rtg,
432 : struct xfs_inode *ip,
433 : struct xbtree_ifakeroot *ifake)
434 : {
435 28374 : struct xfs_btree_cur *cur;
436 :
437 28374 : cur = xfs_rtrefcountbt_init_common(mp, NULL, rtg, ip);
438 28377 : cur->bc_nlevels = ifake->if_levels;
439 28377 : cur->bc_ino.forksize = ifake->if_fork_size;
440 28377 : cur->bc_ino.whichfork = -1;
441 28377 : xfs_btree_stage_ifakeroot(cur, ifake, NULL);
442 28357 : return cur;
443 : }
444 :
445 : /*
446 : * Install a new rt reverse mapping btree root. Caller is responsible for
447 : * invalidating and freeing the old btree blocks.
448 : */
449 : void
450 28340 : xfs_rtrefcountbt_commit_staged_btree(
451 : struct xfs_btree_cur *cur,
452 : struct xfs_trans *tp)
453 : {
454 28340 : struct xbtree_ifakeroot *ifake = cur->bc_ino.ifake;
455 28340 : struct xfs_ifork *ifp;
456 28340 : int flags = XFS_ILOG_CORE | XFS_ILOG_DBROOT;
457 :
458 28340 : ASSERT(cur->bc_flags & XFS_BTREE_STAGING);
459 28340 : ASSERT(ifake->if_fork->if_format == XFS_DINODE_FMT_REFCOUNT);
460 :
461 : /*
462 : * Free any resources hanging off the real fork, then shallow-copy the
463 : * staging fork's contents into the real fork to transfer everything
464 : * we just built.
465 : */
466 28340 : ifp = xfs_ifork_ptr(cur->bc_ino.ip, XFS_DATA_FORK);
467 28340 : xfs_idestroy_fork(ifp);
468 56726 : memcpy(ifp, ifake->if_fork, sizeof(struct xfs_ifork));
469 :
470 28363 : xfs_trans_log_inode(tp, cur->bc_ino.ip, flags);
471 28372 : xfs_btree_commit_ifakeroot(cur, tp, XFS_DATA_FORK,
472 : &xfs_rtrefcountbt_ops);
473 28365 : }
474 :
475 : /* Calculate number of records in a realtime refcount btree block. */
476 : static inline unsigned int
477 : xfs_rtrefcountbt_block_maxrecs(
478 : unsigned int blocklen,
479 : bool leaf)
480 : {
481 :
482 1420258482 : if (leaf)
483 9838420 : return blocklen / sizeof(struct xfs_refcount_rec);
484 1410390521 : return blocklen / (sizeof(struct xfs_refcount_key) +
485 : sizeof(xfs_rtrefcount_ptr_t));
486 : }
487 :
488 : /*
489 : * Calculate number of records in an refcount btree block.
490 : */
491 : unsigned int
492 6504386 : xfs_rtrefcountbt_maxrecs(
493 : struct xfs_mount *mp,
494 : unsigned int blocklen,
495 : bool leaf)
496 : {
497 1420258482 : blocklen -= XFS_RTREFCOUNT_BLOCK_LEN;
498 1420258482 : return xfs_rtrefcountbt_block_maxrecs(blocklen, leaf);
499 : }
500 :
501 : /* Compute the max possible height for realtime refcount btrees. */
502 : unsigned int
503 59 : xfs_rtrefcountbt_maxlevels_ondisk(void)
504 : {
505 59 : unsigned int minrecs[2];
506 59 : unsigned int blocklen;
507 :
508 59 : blocklen = XFS_MIN_CRC_BLOCKSIZE - XFS_BTREE_LBLOCK_CRC_LEN;
509 :
510 59 : minrecs[0] = xfs_rtrefcountbt_block_maxrecs(blocklen, true) / 2;
511 59 : minrecs[1] = xfs_rtrefcountbt_block_maxrecs(blocklen, false) / 2;
512 :
513 : /* We need at most one record for every block in an rt group. */
514 59 : return xfs_btree_compute_maxlevels(minrecs, XFS_MAX_RGBLOCKS);
515 : }
516 :
517 : int __init
518 59 : xfs_rtrefcountbt_init_cur_cache(void)
519 : {
520 59 : xfs_rtrefcountbt_cur_cache = kmem_cache_create("xfs_rtrefcountbt_cur",
521 59 : xfs_btree_cur_sizeof(
522 : xfs_rtrefcountbt_maxlevels_ondisk()),
523 : 0, 0, NULL);
524 :
525 59 : if (!xfs_rtrefcountbt_cur_cache)
526 0 : return -ENOMEM;
527 : return 0;
528 : }
529 :
530 : void
531 58 : xfs_rtrefcountbt_destroy_cur_cache(void)
532 : {
533 58 : kmem_cache_destroy(xfs_rtrefcountbt_cur_cache);
534 58 : xfs_rtrefcountbt_cur_cache = NULL;
535 58 : }
536 :
537 : /* Compute the maximum height of a realtime refcount btree. */
538 : void
539 70799 : xfs_rtrefcountbt_compute_maxlevels(
540 : struct xfs_mount *mp)
541 : {
542 70799 : unsigned int d_maxlevels, r_maxlevels;
543 :
544 70799 : if (!xfs_has_rtreflink(mp)) {
545 50044 : mp->m_rtrefc_maxlevels = 0;
546 50044 : return;
547 : }
548 :
549 : /*
550 : * The realtime refcountbt lives on the data device, which means that
551 : * its maximum height is constrained by the size of the data device and
552 : * the height required to store one refcount record for each rtextent
553 : * in an rt group.
554 : */
555 20755 : d_maxlevels = xfs_btree_space_to_height(mp->m_rtrefc_mnr,
556 : mp->m_sb.sb_dblocks);
557 20755 : r_maxlevels = xfs_btree_compute_maxlevels(mp->m_rtrefc_mnr,
558 20755 : xfs_rtb_to_rtxt(mp, mp->m_sb.sb_rgblocks));
559 :
560 : /* Add one level to handle the inode root level. */
561 20755 : mp->m_rtrefc_maxlevels = min(d_maxlevels, r_maxlevels) + 1;
562 : }
563 :
564 : #define XFS_RTREFC_NAMELEN 21
565 :
566 : /* Create the metadata directory path for an rtrefcount btree inode. */
567 : int
568 100384 : xfs_rtrefcountbt_create_path(
569 : struct xfs_mount *mp,
570 : xfs_rgnumber_t rgno,
571 : struct xfs_imeta_path **pathp)
572 : {
573 100384 : struct xfs_imeta_path *path;
574 100384 : unsigned char *fname;
575 100384 : int error;
576 :
577 100384 : error = xfs_imeta_create_file_path(mp, 2, &path);
578 100384 : if (error)
579 : return error;
580 :
581 100384 : fname = kmalloc(XFS_RTREFC_NAMELEN, GFP_KERNEL);
582 100384 : if (!fname) {
583 0 : xfs_imeta_free_path(path);
584 0 : return -ENOMEM;
585 : }
586 :
587 100384 : snprintf(fname, XFS_RTREFC_NAMELEN, "%u.refcount", rgno);
588 100384 : path->im_path[0] = "realtime";
589 100384 : path->im_path[1] = fname;
590 100384 : path->im_dynamicmask = 0x2;
591 100384 : *pathp = path;
592 100384 : return 0;
593 : }
594 :
595 : /* Calculate the rtrefcount btree size for some records. */
596 : unsigned long long
597 0 : xfs_rtrefcountbt_calc_size(
598 : struct xfs_mount *mp,
599 : unsigned long long len)
600 : {
601 0 : return xfs_btree_calc_size(mp->m_rtrefc_mnr, len);
602 : }
603 :
604 : /*
605 : * Calculate the maximum refcount btree size.
606 : */
607 : static unsigned long long
608 : xfs_rtrefcountbt_max_size(
609 : struct xfs_mount *mp,
610 : xfs_rtblock_t rtblocks)
611 : {
612 : /* Bail out if we're uninitialized, which can happen in mkfs. */
613 198055 : if (mp->m_rtrefc_mxr[0] == 0)
614 : return 0;
615 :
616 198055 : return xfs_rtrefcountbt_calc_size(mp, rtblocks);
617 : }
618 :
619 : /*
620 : * Figure out how many blocks to reserve and how many are used by this btree.
621 : * We need enough space to hold one record for every rt extent in the rtgroup.
622 : */
623 : xfs_filblks_t
624 198325 : xfs_rtrefcountbt_calc_reserves(
625 : struct xfs_mount *mp)
626 : {
627 198325 : if (!xfs_has_rtreflink(mp))
628 : return 0;
629 :
630 198055 : return xfs_rtrefcountbt_max_size(mp,
631 198055 : xfs_rtb_to_rtxt(mp, mp->m_sb.sb_rgblocks));
632 : }
633 :
634 : /*
635 : * Convert on-disk form of btree root to in-memory form.
636 : */
637 : STATIC void
638 97301 : xfs_rtrefcountbt_from_disk(
639 : struct xfs_inode *ip,
640 : struct xfs_rtrefcount_root *dblock,
641 : int dblocklen,
642 : struct xfs_btree_block *rblock)
643 : {
644 97301 : struct xfs_mount *mp = ip->i_mount;
645 97301 : struct xfs_refcount_key *fkp;
646 97301 : __be64 *fpp;
647 97301 : struct xfs_refcount_key *tkp;
648 97301 : __be64 *tpp;
649 97301 : struct xfs_refcount_rec *frp;
650 97301 : struct xfs_refcount_rec *trp;
651 97301 : unsigned int numrecs;
652 97301 : unsigned int maxrecs;
653 97301 : unsigned int rblocklen;
654 :
655 97301 : rblocklen = xfs_rtrefcount_broot_space(mp, dblock);
656 :
657 97301 : xfs_btree_init_block(mp, rblock, &xfs_rtrefcountbt_ops, 0, 0,
658 : ip->i_ino);
659 :
660 97301 : rblock->bb_level = dblock->bb_level;
661 97301 : rblock->bb_numrecs = dblock->bb_numrecs;
662 :
663 97301 : if (be16_to_cpu(rblock->bb_level) > 0) {
664 1601 : maxrecs = xfs_rtrefcountbt_droot_maxrecs(dblocklen, false);
665 1601 : fkp = xfs_rtrefcount_droot_key_addr(dblock, 1);
666 1601 : tkp = xfs_rtrefcount_key_addr(rblock, 1);
667 1601 : fpp = xfs_rtrefcount_droot_ptr_addr(dblock, 1, maxrecs);
668 1601 : tpp = xfs_rtrefcount_broot_ptr_addr(mp, rblock, 1, rblocklen);
669 1601 : numrecs = be16_to_cpu(dblock->bb_numrecs);
670 3202 : memcpy(tkp, fkp, 2 * sizeof(*fkp) * numrecs);
671 3202 : memcpy(tpp, fpp, sizeof(*fpp) * numrecs);
672 : } else {
673 95700 : frp = xfs_rtrefcount_droot_rec_addr(dblock, 1);
674 95700 : trp = xfs_rtrefcount_rec_addr(rblock, 1);
675 95700 : numrecs = be16_to_cpu(dblock->bb_numrecs);
676 191400 : memcpy(trp, frp, sizeof(*frp) * numrecs);
677 : }
678 97301 : }
679 :
680 : /* Load a realtime reference count btree root in from disk. */
681 : int
682 97301 : xfs_iformat_rtrefcount(
683 : struct xfs_inode *ip,
684 : struct xfs_dinode *dip)
685 : {
686 97301 : struct xfs_mount *mp = ip->i_mount;
687 97301 : struct xfs_ifork *ifp = xfs_ifork_ptr(ip, XFS_DATA_FORK);
688 97301 : struct xfs_rtrefcount_root *dfp = XFS_DFORK_PTR(dip, XFS_DATA_FORK);
689 97301 : unsigned int numrecs;
690 97301 : unsigned int level;
691 97301 : int dsize;
692 :
693 97301 : dsize = XFS_DFORK_SIZE(dip, mp, XFS_DATA_FORK);
694 97301 : numrecs = be16_to_cpu(dfp->bb_numrecs);
695 97301 : level = be16_to_cpu(dfp->bb_level);
696 :
697 97301 : if (level > mp->m_rtrefc_maxlevels ||
698 97301 : xfs_rtrefcount_droot_space_calc(level, numrecs) > dsize) {
699 0 : xfs_inode_mark_sick(ip, XFS_SICK_INO_CORE);
700 0 : return -EFSCORRUPTED;
701 : }
702 :
703 194602 : xfs_iroot_alloc(ip, XFS_DATA_FORK,
704 : xfs_rtrefcount_broot_space_calc(mp, level, numrecs));
705 97301 : xfs_rtrefcountbt_from_disk(ip, dfp, dsize, ifp->if_broot);
706 97301 : return 0;
707 : }
708 :
709 : /*
710 : * Convert in-memory form of btree root to on-disk form.
711 : */
712 : void
713 45237 : xfs_rtrefcountbt_to_disk(
714 : struct xfs_mount *mp,
715 : struct xfs_btree_block *rblock,
716 : int rblocklen,
717 : struct xfs_rtrefcount_root *dblock,
718 : int dblocklen)
719 : {
720 45237 : struct xfs_refcount_key *fkp;
721 45237 : __be64 *fpp;
722 45237 : struct xfs_refcount_key *tkp;
723 45237 : __be64 *tpp;
724 45237 : struct xfs_refcount_rec *frp;
725 45237 : struct xfs_refcount_rec *trp;
726 45237 : unsigned int maxrecs;
727 45237 : unsigned int numrecs;
728 :
729 45237 : ASSERT(rblock->bb_magic == cpu_to_be32(XFS_RTREFC_CRC_MAGIC));
730 45237 : ASSERT(uuid_equal(&rblock->bb_u.l.bb_uuid, &mp->m_sb.sb_meta_uuid));
731 45237 : ASSERT(rblock->bb_u.l.bb_blkno == cpu_to_be64(XFS_BUF_DADDR_NULL));
732 45237 : ASSERT(rblock->bb_u.l.bb_leftsib == cpu_to_be64(NULLFSBLOCK));
733 45237 : ASSERT(rblock->bb_u.l.bb_rightsib == cpu_to_be64(NULLFSBLOCK));
734 :
735 45237 : dblock->bb_level = rblock->bb_level;
736 45237 : dblock->bb_numrecs = rblock->bb_numrecs;
737 :
738 45237 : if (be16_to_cpu(rblock->bb_level) > 0) {
739 23292 : maxrecs = xfs_rtrefcountbt_droot_maxrecs(dblocklen, false);
740 23292 : fkp = xfs_rtrefcount_key_addr(rblock, 1);
741 23292 : tkp = xfs_rtrefcount_droot_key_addr(dblock, 1);
742 23292 : fpp = xfs_rtrefcount_broot_ptr_addr(mp, rblock, 1, rblocklen);
743 23292 : tpp = xfs_rtrefcount_droot_ptr_addr(dblock, 1, maxrecs);
744 23292 : numrecs = be16_to_cpu(rblock->bb_numrecs);
745 46584 : memcpy(tkp, fkp, 2 * sizeof(*fkp) * numrecs);
746 46584 : memcpy(tpp, fpp, sizeof(*fpp) * numrecs);
747 : } else {
748 21945 : frp = xfs_rtrefcount_rec_addr(rblock, 1);
749 21945 : trp = xfs_rtrefcount_droot_rec_addr(dblock, 1);
750 21945 : numrecs = be16_to_cpu(rblock->bb_numrecs);
751 43890 : memcpy(trp, frp, sizeof(*frp) * numrecs);
752 : }
753 45237 : }
754 :
755 : /* Flush a realtime reference count btree root out to disk. */
756 : void
757 44743 : xfs_iflush_rtrefcount(
758 : struct xfs_inode *ip,
759 : struct xfs_dinode *dip)
760 : {
761 44743 : struct xfs_ifork *ifp = xfs_ifork_ptr(ip, XFS_DATA_FORK);
762 44743 : struct xfs_rtrefcount_root *dfp = XFS_DFORK_PTR(dip, XFS_DATA_FORK);
763 :
764 44743 : ASSERT(ifp->if_broot != NULL);
765 44743 : ASSERT(ifp->if_broot_bytes > 0);
766 89486 : ASSERT(xfs_rtrefcount_droot_space(ifp->if_broot) <=
767 : xfs_inode_fork_size(ip, XFS_DATA_FORK));
768 134229 : xfs_rtrefcountbt_to_disk(ip->i_mount, ifp->if_broot,
769 44743 : ifp->if_broot_bytes, dfp,
770 44743 : XFS_DFORK_SIZE(dip, ip->i_mount, XFS_DATA_FORK));
771 44743 : }
772 :
773 : /*
774 : * Create a realtime refcount btree inode.
775 : *
776 : * Regardless of the return value, the caller must clean up @upd. If a new
777 : * inode is returned through *ipp, the caller must finish setting up the incore
778 : * inode and release it.
779 : */
780 : int
781 3083 : xfs_rtrefcountbt_create(
782 : struct xfs_imeta_update *upd,
783 : struct xfs_inode **ipp)
784 : {
785 3083 : struct xfs_mount *mp = upd->mp;
786 3083 : struct xfs_ifork *ifp;
787 3083 : int error;
788 :
789 3083 : error = xfs_imeta_create(upd, S_IFREG, ipp);
790 3083 : if (error)
791 : return error;
792 :
793 3083 : ifp = xfs_ifork_ptr(upd->ip, XFS_DATA_FORK);
794 3083 : ifp->if_format = XFS_DINODE_FMT_REFCOUNT;
795 3083 : ASSERT(ifp->if_broot_bytes == 0);
796 3083 : ASSERT(ifp->if_bytes == 0);
797 :
798 : /* Initialize the empty incore btree root. */
799 3083 : xfs_iroot_alloc(upd->ip, XFS_DATA_FORK,
800 : xfs_rtrefcount_broot_space_calc(mp, 0, 0));
801 3083 : xfs_btree_init_block(mp, ifp->if_broot, &xfs_rtrefcountbt_ops,
802 3083 : 0, 0, upd->ip->i_ino);
803 3083 : xfs_trans_log_inode(upd->tp, upd->ip, XFS_ILOG_CORE | XFS_ILOG_DBROOT);
804 3083 : return 0;
805 : }
|
